ON HOPF BIFURCATION OF RAYLEIGH'S EQUATION M. FARKAS, L. SPARING and Gy. SZAB6
Mechanical Engineering, Department of Mathematics.
Technical University. H-1521 Budapest, Received August 28. 1984
Summary
Rayleigh's differential equation exhibits a supercritical Hopfbifurcation. The bifurcating closed orbits correspond to fixed points of the Poincare map. In this paper explicit estimates of the domain of the Poincare map are given and these estimates are checked by a computer. The results may help the estimate of the possible amplitude of the stable oscillation of this important nonlinear system.
Introduction
Rayleigh's differential equation
d
2
U+dU((du))"2 1
1)+11=0, pER
dt
2dt dt
(1)is an important model used extensively in the theory of non-linear oscillations (see e. g. Minorsky [3J). A system equivalent to the second order scalar equation (I) is
(2) where dot denotes differentiation with respect to t ER. This system satisfies all the conditions of the classical Hopf bifurcation theorem [1
J
(or see Marsden- McCracken [2J). That is to say, the origin x=(x1 , x2)=0 is an equilibrium of system (2) for all 1I E R. For 1I <°
it is asymptotically stable, for p >°
it isunstable. A 6>
°
and a smooth function 11: ( - 6, (»)f----7 R exist such that prO)=
0,1/(0)
= °
and for all x? E ( - (), ()) the solution ofcorresponding to the initial condition (x?, 0) is periodic, and in a three dimensional neighbourhood of (x l' X 2' It) = (0,0,0) the corresponding trajec- tories arc the only closed paths of system (2). Moreover. applying the method of
264 .If. FARKAS (.[ af.
Negrini-Salvadori [4J a Lyapunov function
2 2 3 3 5 3
F(X 1,X2)=X 1 +x2+ 4X1X2+ 4 X1X2
can be determined which is, obviously, positive definite in a neighbourhood of the origin x
= °
and its derivative with respect to system (2) with It= °
isnegative definite: 3
F(x1 , X 2 ) = - 4 (x~
+
X~)2+
e(lxI4).This implies that x =
°
is a 3-attractor (a vague attractor) of system (2) with fJ.= °
and, as a consequence, fJ.(x 1) > 0, Xl :;6 0, and the bifurcating closed paths are orbitally asymptotically stable.In this paper we are giving an explicit extimate of the domain of the Poincare map attached to the problem. This estimate may enable us to estimate either the domain or the range of the function fJ., i. e. the values of the bifurcation parameter fJ. and the initial conditions x? for which still bifurcating closed paths exist. (It is, of course, known that (1) has a non-constant periodic solution for all fJ.:;6 0, since, (1) is equivalent to van der Pol's equation if fJ.:;6 0). Our method seems to be fairly general and possibly applicable in treating bifurcations of other two dimensional sy<;tems. However, the estimates gained by this exact method are rather c( nservative compared to results gained by computer experiments. In section 2 the domain of the Poincare map is estimated from below, and in section 3 the results gained by a computer are presented.
The Estimate of the Poincare Map
First of all, system (2) will be transformed by the polar transtormation
Xl =r cos
e,
X 2 =r sine
intor=r
sin2 e(fJ.-r2 sin2 e)e =
-1- r2 2 sin 2e sin2 e+ I
sin 2e. (3)The solution of (3) assuming the initial values (r, e) at t
= °
will be denoted by (f(t, r, e, fJ.), 8(t, r, e, fJ.)). It will be assumed always that r"2 0. If the initial values are (0, e) then, clearly, f(t, 0, e, fJ.)=0 and the function 8(t, 0, e, fJ.) satisfies.:.. fJ.. -
e(t, 0, e, fJ.)= -1
+ "2
sIn2e(t, 0, e, fJ.). (4) In particular, 8(t, 0, 0, 0) = - t and 8(21[,0,0,0) = - 21[. As a consequence, it is clear that for small enough fJ. and Initial condition (r, 0) the integral curves of the265
solutions of (3) will cut the
e
= - 2n plane (in t, r,e
space) at some momentt = T(p, r) > 0, see Fig. 1.
I n view of the polar transformation the initial .::ondition (x l' x 2)
=
(r, 0)attached to system (2) corresponds to the initial condition (r, 0) attached to (3).
21[ T(.u. r )
---
---
Fig. I. Integral curves of the solutions of (3) ::>:: f.L=0, (r, 8)=(0,0);
fJ: f.L smalL r small, 8 =
°
If (<P1(t, X 1 ,X 2,P), <P2(t, Xl' x2 , p)) denotes the solution of (2) satisfying the initial conditions
(<p 1 (0, Xl' X 2' Il), <P2 (0, Xl' X 2, p)) = (X 1 , x 2) then
<PI (T(p, r), r, 0, p)=f(T(p, r), r, 0, p) cos (-2n)=
=
f(T(p, r), r, 0, p) 2:: 0,<P2(T(p, r), r, 0, p)=f(T(p, r), r, 0, p) sin (-2n)=0.
This means that the path corresponding to the solution (<Pl(t, r, 0, p), <P2(t, r, 0, p)) of (2) intersects the positive Xl axis at moment t= T(p, r) the first time after the intersection (r, 0) at t
=
0. See Fig. 2. The Poincare map, the domain of which we want to estimate, is the mapping defined by(p, r)f--+ <P 1 (T(p, r), r, 0, p). (5) In order to get an estimate for the domain of this mapping we have to establish an estimate for the solutions of (3).
266 If. f-"AR};AS "f <1/.
LEMMA. The solution (f(t, r, 0, fl), 8(t, r, 0, fl)) of (3) corresponding to the initial condition (r, 0) r> 0 and to the parameter value J-i:2: 0 is defined in
Os t < 00, and satisfies the inequality
O<f(t, r, 0, fl)sr(l +tfl2/2r2)1/2. (6)
x,
X,
o
Fig, 2. Poincare map attached to (2) around the origin
PROOF. If r > 0 the corresponding path cannot cross the axis r
=
0 since the latter is itself a trajectory, thus, f(t, r, 0, fl) > 0 in its existence domain. Taking this into account we estimate the maximum of the function standing on the right hand side of the first equation of (3) for fixed r > 0 and fl ~ 0:After a somewhat lengthy but elementary calculation we get
Here the right hand side is the actual value of the maximum if 0 S fl S 2r2 and it is an upper estimate in case J1 > 2r2. Thus, the following inequality holds
Solving this differential inequality we get (6) for t? 0 in the existence domain of the solution. The right hand side of (6) is bounded in every bounded interval.
Thus, f is bounded in every bounded interval, and in view of the second
ON HOPF BIFURCATION OF RA YLElGHS EQUATlOS 267
equation of(3), the same is true for
19
and, as a consequence, fore.
This implies that the solution is defined in [0, CI)), and this completes the proof.As it was shown at the beginning of this section, the integral curve of (3) corresponding to the initial condition (0, 0) and to the parameter value J1. =
°
cuts the 8
= -
2n plane at t=
2n. Therefore, we are going to fix an interval larger than [0, 2n] to ensure a crossing of the same plane by the integral curve corresponding to the initial condition (r, 0), r:?:°
and the parameter value J1.:?: 0.THEOREM. Let us choose a constant a> 1 and let J1. and r satisfy the inequalities
(7) then the integral curve of the solution (1';(t, r, 0, J1.), e(t, r, 0, J1.)) cuts the plane 8 =
- 2n at some moment T(J1., r) E (0, 2an].
PROOF. We get from the second inequality of (7) that
for t E [0, 2an], r > 0. Hence applying (6) we get that p2(t, r, 0, J1.) ~ 2(a - 1 )/a
for t E [0, 2an], r > 0, J1.:?: 0. However, the last inequality is trivially true also for r 0. The first inequality of(7) and the one we have just obtained imply (the writing out of the arguments will be omitted)
.:.. 1 ? ? _ _
8(t, r. 0, J1.)= -1
+
2 (p-r- sin-8) sin 28 ~1 1
1
- ? ?-I
a-I 1~
- +?
p-r-sm- 8 ~ -1+ - - = - - .
- a a
Integrating from
°
to 2an we get e(2an, r, 0, /-l) ~ - 2n. Thus, for some 0< T(J1., r) ~ 2an we have e(T(J1., r), r, 0, p) = - 2n and this was to be proved.COROLLARY. For any fixed a> 1 the set Da defined by (7) is a subset of the domain of the Poincare map (5).
268 .If. FARKAS <', ul.
Numerical calculations show that the optimal set Da is obtained if a = 2.5 is chosen. Introducing the notation
Fa(Jl)=(2(a-l)/a- anp2)1/2
Figure 3 shows the set D2 . 5 and the graph of the function Jl discussed in the Introduction. The latter has been gained by numerical methods.
J
OL-____________ ~~ ______ ~~ __ ~~ ____ ~
032 0.39 j.J
0.1
Fil!. 3. Estimate of the domain of the Poincare map
Numerical Results
The fixed points of the Poincare map defined in (5) can be determined by computer. To each r> 0 small enough the value p (r) of the function p can be uniquely determined for which r= CfJdT(p(r), r), r, 0, p(r)) holds. The corresponding solution (CfJl(t, r, 0, p(r)), CfJ2(t, r, 0, p(r)) of(2) (where p=Jl(r)) is periodic with period T(p(r), r).
We have used a second generation computer of type ODRA-1204 and a elL digital plotter to get graphical results. A Runge-Kutta type method was applied to solve system (2). We could bring down the relative error of the Poincare map below 10 -5, keeping, at the same time, computer time within reasonable limits. We have determined the values of the function p and the corresponding periods T(p(r), r) first within the limits of the domain established
--- - - - - -_ .. _----_._-_. -_.-._ ... -
O,V 1/01'1' BII'( ReAT/OS OF RA ),I.U(;IIS ICQLITlOS 269
theoretically (Fig. 3) then the values of r were increased considerably and the closed paths remained to exist, though became more and more deformed. The results are shown in Table I. It can be seen from the fourth column that for small values of r we have Il(r):::::0.75r2.
Table 1
Initial value (r). corresponding bifurcation parameter value II(r), period T(IL(r), r) and II(r)/r2 for Rayleigh's equation
/£(r) T(/I(r), r) p(r)/r2
.1 .om 503 6.283207 0.7503
.2 .029999 6.283538 0.7500
J .067451 6.2R4970 0.7495
.4 .119699 6.288809 0.7481
.5 .186359 6.296 ROR 0.7454
.6 .266640 6.311050 0.7407
.7 .359269 6.333690 0.7332
.8 .462407 6.366648 0.7225
.9 .573786 6.41 1260 0.7084
1.0 .690961 6.468034 0.6910
LI .81 I 566 6.536639 0.6707
1.2 (.J33595 6.616046 0.6483
L3 L05549X 6.704795 0.6246
1.4 L176237 6.801289 0.6001
1.5 1.295205 6.903970 0.5756
2.0 1.857 I11 7.469813 0.4643
2.5 2.370079 X.065658 0.3792
3.0 2.846108 8.660461 0.3162
3.5 3.293772 9.2451XO 0.2689
4.0 3.718595 9.817202 0.2324
4.5 4.124565 10.376058 0.2037
5.0 4.514543 10.922108 0.1806
6.0 5.254925 I L97X 473 0.1460
7.0 5.952409 12.992 099 0.1215
8.0 6.615321 13.968404 0.1034
9.0 7.249579 14.911949 0.0895
lO.O 7.X59517 15.826517 0.07R6
Figure 4 shows some of the bifurcating closed paths from r = 0.1 to r = 1.5.
Figure 5 shows the graphs of the functions n--;. tl(r) and rf--7 T(fl(r), r). One can see from Figure 6 how strongly attractive is the closed path corresponding to r = 1, fl{ 1 ) = 0.69096 yet.
270 \I. FA RKA,,)' et ai.
x,
Fill. 4. Bifurcating closed paths of Rayleigh's equation
t Y 10
15 l=Tlplrl.rl
10 r Fill. 5. Graphs of the functions n--+ p(r) and n ... T(p(r).r)
ON HOP!' Bln"RCATlON OF RA )'LE/GfI'S EQUATION 271
X,
Fig. 6. The attractive closed path of Rayleigh's equation for J1 = 0.69096
References
1. HOPF, E.: Abzweigung einer periodischen Losung von einer stationaren Losung eines Differentialsystems, Ber. Math.-Phys. Sachsische Akad. Wiss. Leipzig, 94 I. (1942).
2. MARSDEN,1. E.-McCRACKEN, M.: The Hopfbifurcation and its applications, Springer, New York, 1976.
3. MINORSKY, N.: Introduction to non-linear mechanics, Edwards, Ann Arbor, 1947.
4. NEGR!N!, P.-SALVADOR!, L.: Attractivity and Hopf bifurcation, Nonlin. Anal. T. M. A. 3 87.
(1979)
Prof. Dr. Miklos F ARKAS
I
L<iszlo SPARING, H-1521, Budapest Dr. Gyorgy Szabo